ARTICLE IN PRESS

JOURNAL OF
SOUND AND
VIBRATION
Journal of Sound and Vibration 300 (2007) 744–762
www.elsevier.com/locate/jsvi

The measurement of the loss factors of beams and

plates with constrained and unconstrained
damping layers: A critical assessment
Denys J. Mead
Institute of Sound and Vibration Research, University of Southampton, Southampton SO17 1BJ, UK
Received 24 February 2006; received in revised form 22 August 2006; accepted 23 August 2006
Available online 17 October 2006

Abstract

Methods of measuring the loss factors of heavily damped beams and plates damped by uniform layers of visco-elastic
damping are reviewed and compared. Deducing loss factors from measured spatial decay rates of free flexural waves in
long beams is shown to be valid only for wave motion governed by the fourth-order flexural wave equation, so the method
can only be used for unconstrained layer configurations. The usual approximation made in this method limits its accuracy
and is examined. The flexural wave equation for three-layered sandwich beams and plates with constrained damping layers
is of sixth order, so the spatial decay rate method is inapplicable for deducing loss factors. This is shown by both analysis
and computer experiment. Reliable methods of damping measurement on finite damped sandwich beams are then
investigated by computer experiment. Frequency-response functions for three-layered sandwich beams are computed for
free–free and encastré beams, centrally excited. Loss factors of the fundamental symmetric modes are deduced from this
computed response data as though it were experimental data. They are deduced by established methods such as the ‘half-
power-point’ method, detailed analysis of Nyquist diagrams and the energy input method. Most of the deduced loss factors
agree closely with the loss factors of the theoretical damped normal modes.
r 2006 Elsevier Ltd. All rights reserved.

1. Introduction

Experimental measurement of the loss factors of heavily damped beams and plates became necessary in the
early 1950s when the flexural vibration of plate and beam-like structures was first damped by the newly
developed and efficient polymeric damping materials. The earliest materials were simply sprayed or trowelled
on the structural surfaces in layers, the whole system then becoming the so-called ‘unconstrained damping
layer’ configuration. More effective is the ‘constrained layer’ configuration in which the damping material is
bonded and sandwiched between two plates or beams and constrained to undergo energy-dissipating shear
strain.
The system damping had to be measured over as wide a frequency and temperature range as possible.
A suitable method devised for unconstrained layer configurations involved a long and narrow damped beam,

E-mail address: [email protected].

0022-460X/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jsv.2006.08.023
 ARTICLE IN PRESS
D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762 745

Nomenclature M mass of a single-degree of freedom

system
E1, E3 Young’s moduli of the outer plates of the PEl maximum potential energy stored in one
sandwich wavelength of the beam
EI real part of the complex flexural stiffness Q the Q-factor, response amplitude at
Etotal total maximum energy of vibration in resonance/response at zero frequency
one wavelength of a beam T(o) complex transmissibility of vibration at
EIT total flexural stiffness of the two outer frequency o from the source to the
plates of a sandwich receiver point
F amplitude of harmonic force acting on a w, w(x, t) (complex) transverse displacement of
single-degree of freedom system beam
g shear parameter X the quotient Re(w)/Im(w)
G shear modulus of the damping layer Y geometric parameter
h1, h2, h3 thickness of outer plate 1, damping layer b loss factor of damping layer material
(layer 2), outer plate 3 d exponential decay rate per wavelength
H hysteretic damping coefficient of a single- e phase difference between motions at two
degree of freedom system points
k complex wavenumber Z loss factor
kr, ki real and imaginary parts of k l wavelength
K stiffness of a single-degree of freedom o radian frequency
system on natural frequency
L a prescribed standard length O non-dimensional frequency
m mass per unit length

excited harmonically at one end while the other end was made as non-reflective as possible. Well away from
either end the predominant flexural wave motion in the beam was therefore that of a single flexural wave
travelling away from the external forcing and decaying as it progressed. Its decay rate per wavelength was
measured from which the flexural loss factor of the beam was deduced.
This method is quite legitimate for the unconstrained configuration. Oberst and Frankenfeld [1] had shown
that when the two-layered configuration vibrates harmonically, its flexural rigidity could be regarded as being
complex, its normalised imaginary part being the flexural loss factor, which is to be measured. Harmonic
bending of such a beam is governed by the simple Euler–Bernoulli theory of bending (the ‘EBT’ theory) which
leads to the familiar fourth-order differential equation for free flexural wave motion. The wavenumbers and
decay rates of the wave motion for any given frequency are easily found from this and a simple relationship
exists between the decay rate and the flexural loss factor. The loss factor is therefore easily found from the
decay rate (and vice versa), so providing the legitimate basis for the ‘long-beam method’ of measuring the
flexural loss factor of a beam with an unconstrained damping layer.
The same cannot be said of the constrained layer-damping configuration. Ross, Ungar and Kerwin [2,3]
were the first to analyse this configuration and (like Oberst) sought its effective complex flexural stiffness. They
found it as a function of the wavenumber and the properties of the cross-section which were contained in three
important non-dimensional parameters. The wavenumber, of course, is a function of the flexural stiffness
which, being a function of the wavenumber, suggests a vicious circle. They escaped it by assuming the spatial
wave motion to be purely sinusoidal, i.e. to have no spatial decay. By specifying the wavelength of the motion
(and hence the wavenumber) they found a corresponding flexural stiffness, loss factor and frequency of wave
propagation.
Now spatially non-decaying sinusoidal motion can only exist if an external distributed harmonic loading
acts on the system to prevent the inevitable spatial decay caused by energy dissipation in the damping layer.
Such an ‘external distributed loading’ is an essential feature of the ‘damped forced normal modes’ introduced
later on by DiTaranto, Mead and Markus [4,5]. The expression for the loss factor derived by Kerwin et al. is
 ARTICLE IN PRESS
746 D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762

quite correct within this context, but their use of the term ‘effective flexural stiffness’ has unfortunately misled
some experimenters. They assumed it implied that the wave motion in a constrained layer configuration could
be described by a fourth-order differential wave equation, as for the unconstrained layer beam. If that were
true, the former relationship between the wave decay rate and the loss factor could be used to deduce loss
factors from measured decay rates. Part of the purpose of this paper is to emphasise and demonstrate the error
of this assumption and to compare some of the erroneous results it produces with the true values.
Wave motion in a three-layered constrained-layer damping configuration is actually governed by a sixth-
order differential equation. First derived by Mead and Markus [6], it was then used by DiTaranto, Mead and
Markus [4,5] to find the loss factors of finite sandwich beams and plates with arbitrary boundary conditions.
Both their theory (the ‘M&M’ theory) and the sixth-order equation incorporate the same non-dimensional
parameters as in Refs. [2,3], but an effective complex flexural modulus is no longer sought. A simple formula
for the loss factor cannot be derived and the value of the loss factor can only be found by an iterative
computational process.
The sixth-order equation and its solution will be used in this paper for three purposes:

(i) to find the complex wavenumbers of spatially decaying free harmonic waves in unbounded damped
sandwich beams at any real frequency,
(ii) to find the complex frequencies and the true loss factors of the damped normal modes of finite sandwich
beams with arbitrary boundary conditions,
(iii) to find the complex frequency response of finite damped sandwich beams with specific boundary
conditions, excited by a single-point harmonic force.

From (i), it is easily shown that six spatially decaying flexural waves can exist at any real frequency, their
wavenumbers occurring in three equal and opposite pairs. One of the pairs has a much smaller decay rate than
the other two, so one of these waves necessarily dominates the motion in a constrained layered beam in the
‘long beam’ experiment already described. Its decay rate can be measured in the laboratory or can be
computed at the desk. Part of the purpose of this paper is to investigate by ‘computer experiment’ whether
these decay rates can be used to deduce the true loss factors of the beam.
From the complex frequencies1 of (ii) the ‘true loss factors ‘ may be found. The ‘true loss factor’ must be
carefully defined and is best expressed in terms of an energy ratio to be quoted later (see also Ref. [7]). The
M&M theory determines the complex frequencies of the complex ‘damped normal modes’ and identifies the
normalised imaginary parts of these frequencies with the true loss factors. These have been shown to satisfy the
energy definition of the loss factor although the M&M theory does not use this definition to compute them [8].
Now the damped normal modes of finite sandwich beams or plates cannot be purely excited in normal
laboratory practice and they can only be excited in their purest form by their own unique distributions of
external loading. These can never be simulated in practice. The usual (and simplest) practical form of
excitation is by a single-point harmonic force of variable frequency. At a resonance forced in this way the
response of the damped sandwich consists of a superposition of all the infinite number of damped normal
modes (a well-known feature of even lightly damped structures) and it leads to problems of modal
identification and loss factor measurement when resonance frequencies are closely spaced. However, well-
established methods have been developed for analysing measured frequency-response functions to extract
modal loss factors. Which of these methods, when used on heavily damped sandwich beams, yields similar (or
identical) loss factors to the true loss factors needs to be ascertained, and is the other principal feature
investigated in this paper by computer experiment. Exact frequency-response functions are computed for
single-point excited sandwich beams with free–free or fully fixed (encastré) boundary conditions. Exact closed-
form solutions of the sixth-order equation which satisfy the appropriate beam boundary conditions are used
for this. The computed response functions are analysed to yield ‘pseudo-experimental’ loss factors to be
compared with both the true computed loss factors of the damped normal modes and those deduced from
decay rates in long beams, as predicted by computer experiment.
1
The complex frequency concept must not be taken to imply any time-decay of the motion. The imaginary part actually represents the
externally applied distributed loading required to maintain harmonic time-undecaying motion at a resonance frequency.
 ARTICLE IN PRESS
D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762 747

Mention must be made at this point of an excellent experimental and theoretical investigation undertaken
by Lu and Douglas [9]. Basing their work on the sixth-order wave equation and carefully using the correct
frequency-dependent material properties of the damping layer, they computed the frequency-response
function of a freely supported damped sandwich beam over a frequency range extending beyond the fourth
resonance. The experimental frequency-response function agreed excellently with the calculations, but no
investigation was made into modal loss factors.

2. Spatial decay rates of free propagating wave motion in damped beams

2.1. Beams with unconstrained layers of damping material

When the damping material has linear elastic-hysteretic properties, the mode of harmonic flexural
displacement of the beam w(x) is governed by the fourth-order equation
EIð1 þ iZÞw0000  mo2 w ¼ 0. (1)
w0000 implies the fourth x-wise derivative of w. EIð1 þ iZÞ is the complex flexural stiffness of the whole section
of the layered beam, m is the total mass per unit length of the beam and o is the (radian) frequency. The
complex stiffness is given by the expression first derived by Oberst and Frankenfeld [1] and involves the
thicknesses and material properties of the two-layered beam. Further details of this and of other ways of
deriving it are described elsewhere [10].
With the wave displacement expressed in the form wðxÞ ¼ A expðkxÞeiot , Eq. (1) yields four complex values
for k given by ðmo2 =ðEIð1 þ iZÞÞÞ1=4 . Two of these represent wave motion decaying from left to right, while the
other two decay from right to left. When a long damped beam is excited at its left-hand end, the first two of
these are generated at that end but, as one of them decays much more rapidly than the other, its amplitude is
negligible remote from the excited end. The other two waves are generated by reflections from any
discontinuities or boundary to the right of the source, but if they are far enough away from the source or if the
boundary is non-reflective, their magnitudes and presence can also be neglected. In the intermediate region of
the beam the only significant motion is then that of a single wave decaying from left to right. Denote its
complex wavenumber by kr þ iki so that
 1=4
mo2
kr þ iki ¼ . (2)
EIð1 þ iZÞ
Of the four possible values of this, the wave with the lowest decay rate has the (kr, ki) values
 1=4  1=4
mo2 1=4 mo2
ki ¼ Reð1  iZÞ and k r ¼ Imð1  iZÞ1=4 , (3a,b)
EIð1  Z2 Þ EIð1  Z2 Þ
ki here is the usual ‘wavenumber’ which is numerically equal to 2p=l, l being the wavelength of the motion
which is relatively easy to measure on a vibrating beam. ki should also be regarded as the phase difference
between the motion of the beam at two locations along the beam separated by unit distance. The phase
difference e between two points separated by a distance x can be measured, so the wavenumber is e/x and the
wavelength is l ¼ 2px=.
Now kr is the exponential decay rate per unit length along the beam. The more easily measured quantity is
the decay rate per wavelength , i.e. krl. Denote this by d, so
kr
d ¼ kr l ¼ 2p. (4)
ki
To deduce EI and Z from experimentally measured values of ki and d, one proceeds as follows. Rationalising
Eq. (2) one obtains
" #
4
mo2 2 ðk r  iki Þ
EIð1 þ iZÞ ¼ ¼ mo ,
ðkr þ iki Þ4 ðk2r þ k2i Þ4
 ARTICLE IN PRESS
748 D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762

so
mo2 mo2
EI ¼ Reðkr þ iki Þ4 and EIZ ¼ Imðkr þ iki Þ4 .
ðk2rþ k2i Þ4 ðkr þ iki Þ4
Expanding the bracketed functions of kr and ki and substituting d=2p for kr =ki and 2p=l for ki one obtains
 4 
ðd  24d2 p2 þ 16p4 Þ
EI ¼ mo2 l4 (5)
ðd4  4p2 Þ4
and
8pdð4p2  d2 Þ
Z¼ (6)
ðd4  24d2 p2 þ 16p4 Þ
 
If d is small d2  ð2pÞ2 this becomes
Z  2d=p, (7)
which is the approximate expression usually used. It can also be expressed in the form Z ¼ 4kr =ki .
Using the exact Eq. (6) one can examine the range of Z over which this approximation is valid and Fig. 1
compares the exact and approximate values for values of Z up to 1.0. The approximate values are evidently
within 1% of the exact values if Zo0:35, within 7% if Zo0:5 and within 20% if Zo1:0. However, Z values even
as high as 0.2 are seldom realised in usable unconstrained-layer configurations, and spatial decay rates and
wavelengths cannot easily be measured to commensurable degrees of accuracy.

2.2. Beams with constrained layers of damping material

This section considers the three-layer damping configuration in which the damping layer (the ‘core’) is
sandwiched between two elastic layers (‘face-plates’). The face-plates carry both bending stress and transverse
shear stress but with negligible shear strain. Their elastic moduli (Young’s modulus) are E1 and E3,
respectively, and their thicknesses are h1 and h3. It will be assumed (as in Refs. [2–6]) that the core carries
transverse shear stress and undergoes shear strain but has no significant longitudinal direct stress. Its complex
shear modulus is Gð1 þ ibÞ and its thickness is h2. It undergoes no transverse direct strain so all points on a
transverse cross-section of the beam undergo the same transverse deflection w. Under these conditions the

Fig. 1. Approximate loss factors deduced from the decay rates of flexural waves in a long damped Euler–Bernoulli beam and comparison
with the exact values. (————) approximate values, (— — —) exact values, (– – – – – –) quotient, i.e. approximate C exact.
 ARTICLE IN PRESS
D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762 749

differential wave equation of harmonic motion w is found to be

 
EI T wvi  gð1 þ ibÞð1 þ Y Þwiv  o2 m w00  gð1 þ ibÞw ¼ 0 (8)
EI T ¼ ðE 1 h31 E 3 h33 Þ=12 is the sum of the flexural stiffnesses of the face-plates acting independently. g is a
‘shear parameter’ proportional to G and defined in the Appendix. Y (also defined in the Appendix) is a
geometric-cum-flexural stiffness parameter, which depends only upon the face-plate thicknesses and moduli
and the core thickness.
Introduce the non-dimensional frequency O defined by
O2 ¼ o2 mL4 =EI T , (9)
where L is a length to be prescribed later. Set wðxÞ ¼ Aekx so that Eq. (8) yields the following sixth-order (bi-
cubic) polynomial equation for k:
ðkLÞ6  gð1 þ ibÞð1 þ Y ÞðkLÞ4  O2 ðkLÞ2 þ O2 gð1 þ ibÞ ¼ 0. (10)
This has three pairs of complex roots, each of which corresponds to a flexural travelling wave which decays
as it progresses. One of the pair with the lowest decay rates dominates the motion at locations in a long beam
remote from a single-point harmonic exciting force and from a boundary. In principle, if the wavelength and
decay rate could be measured, Eq. (10) could be used to deduce the core shear modulus G and loss factor b but
this is found to be an exceedingly ill-conditioned and impracticable process. In any case, there are much better
ways of measuring the material properties. Notice that Eq. (10) as it stands gives no indication of the loss
factor of the whole beam, whatever that may mean.
It is instructive, however, firstly to consider how the three pairs of k vary with frequency and shear
parameter. Once again, the k’s have the complex form k ¼ kr þ iki and Figs. 2a–c show how their real and
imaginary parts (the decay index and wavenumber, respectively) and their moduli vary with frequency for a

Fig. 2. The complex wave numbers of free wave motion in a damped three-layered sandwich beam: (a) real and imaginary parts, (b) the
moduli, (c) decay rates per wavelength. g ¼ 1, Y ¼ 3:162, b ¼ 1; in (a) — — decay indices (the real parts); in (b) — — — O1=2 .
 ARTICLE IN PRESS
750 D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762
pffiffiffiffiffi
sandwich configuration with g ¼ 1; Y ¼ 3:162 ð¼ 10Þ, b ¼ 1, values which correspond to a very soft, thin
and highly damped sandwiched layer. Each of the three sets of k’s varies with O in a distinctly
pffiffiffiffi different way,
unlike those of the k’s for the damped Euler p beam
ffiffiffiffi which always vary in proportion to O . However, two of
the sandwich sets are almost proportional
pffiffiffiffi to O when O440. This is seen on Fig. 2b, which shows the moduli
of each set and compares them with pffiffiffi Offi .
Now the two sets which approach O are those with the highest decay rates. The set of greatest interest is
the one, which might be detected and measured in a ‘long beam’ experiment, i.e. the one with the lowest decay
rate. Its decay rate reaches a maximum at a frequency of approximately 10, demonstrating the well-known
feature of constrained layer damping that their damping properties reach peak values at specific frequencies
governed by the values of g, Y and b.
The decay indices of Fig. 2a are the decay rates per unit length. The decay rates per wavelength of each wave
motion (d ¼ ðkr =ki Þ2p, Eq. (4)) are shown in Fig. 2c. The lowest curve indicates even more clearly the
characteristic variation of the damping with frequency. Such a curve as this might be generated from
experimentally measured values, provided the motion is large enough to be measured accurately. It still
remains, however, to deduce loss factors from such values. The formula relating loss factors to decay rates of
damped Euler beams (Eq. (6)), being derived from the fourth-order wave equation, may be completely
inapplicable to the sandwich beam, so the following alternative method will be investigated.
As already stated, the system loss factor can be defined in terms of an energy ratio as follows [7,8]:
1 Total energy dissipated in the system per harmonic cycle
Loss factor; Z ¼ . (11)
2p maximum potential energy stored during the cycle
It must be recognised that the maximum energy stored in different elements and locations in the damped
system occurs, in general, at different times in the course of the cycle [8]. For example, the maximum energy in
the core of the damped sandwich does not occur at the same time as the maximum energy in a face-plate and
this must be taken into account when estimating the total maximum energy. Practical difficulties, however,
preclude the use of this expression as a basis for the experimental determination of Z. Instead one must depend
upon the approximate equality of the maximum potential and maximum kinetic energies of the wave motion
when the damping is small. They are exactly equal in each elemental length of both Euler and sandwich beams
in the absence of damping but they become increasingly unequal as the damping increases in sandwich beams.
This will be seen to limit the usefulness of the method.
Denote the total maximum energy (kinetic+potential) in one wavelength of the beam by Etotal. This is
proportional to the square of the displacement amplitude jwðx0 Þj2 at an arbitrary point x0 within that
wavelength, so E total ¼ Cjwðx0 Þj2 . (C is simply the constant of proportionality). Over the next (adjacent)
wavelength, the displacement amplitude at the corresponding point is reduced to jwðxÞjekr l ¼ jwðxÞjed so the
total maximum energy is reduced to CjwðxÞj2 e2d . The difference between these energies is dE l ¼ jwðxÞj2 Cð1 
e2d Þ which is the energy dissipated by the damping mechanism as the wave advances over the spatial cycle of
one wavelength.2 On the assumption that the total maximum potential energy is one-half of the total
maximum energy, the maximum potential energy PEl in the first wavelength is Cjwðx0 Þj2 =2. Now substitute
these expressions for PEl and dEl into the energy Eq. (11) for the loss factor to yield
1 dE l ð1  e2d Þ
Z¼ ¼ . (12)
2p PEl p
The simple first-order approximation for e2d in this leads to Zapprox ¼ 2d=p when d is small, which is the
same as given by Eq. (11).
Clearly, if d becomes very large, Z as given by the Eq. (12) approaches the obviously false value of 1/p. The
error stems from the actual inequality of the maximum kinetic and potential energies in a damped wave,
especially when the decay rates are large. This feature applies particularly to the two waves of Fig. 2 with the
higher decay rates and d’s. These are actually damped evanescent waves. It is well known that, even in the
absence of damping, the potential and kinetic energies in evanescent waves are unequal. The difference
between their maxima is balanced by reactive power from the source of excitation.
2
It is the spatial equivalent of the energy dissipated in a temporal harmonic cycle as required in the energy definition of the loss factor.
 ARTICLE IN PRESS
D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762 751

It must be emphasised that the loss factors deduced in this way pertain to damped waves propagating freely
away from a harmonic point-source. They do not necessarily apply to the constituent waves of resonating
finite sandwich beams, a feature which will be explained and demonstrated later in this paper.

3. The loss factors of single-span simply supported sandwich beams

Ross, Ungar and Kerwin [2,3] assumed the transverse beam displacement varied with x and t in the simple
sinusoidal form wðx; tÞ ¼ w0 eiki x , i.e. with no spatial or temporal decay. This can only be achieved if some form
of external distributed forcing acts on the beam, otherwise the wave must inevitably decay as it propagates.
This external forcing characterises the ‘damped normal modes’ of DiTaranto, Mead and Markus [4,5], which
will be considered in the next section. For the time being, it can be stated that the damped normal mode of a
simply supported damped sandwich beam actually has the simple sinusoidal form given by the above
expression, with ki having a discrete value 2pn=L (n ¼ 1; 2; 3; . . . ; etc. L is the beam length). Substituting this
into Eq. (10) it yields, after rearrangement
ðki LÞ6  gð1 þ ibÞð1 þ Y Þðki LÞ4
O2 ¼ . (13)
ðki LÞ2 þ gð1 þ ibÞ
The presence of the complex term ð1 þ ibÞ on the right-hand side means that O2 is complex and can be
expressed in the form O2 ð1 þ iZÞ. The imaginary part of this is directly related to the distributed forcing of the
mode [4,5] and Z has also been shown [8] to be the genuine loss factor of the damped forced mode as it satisfies
the energy Eq. (11). The distributed forcing is of a particular kind, being equal to Z times the inertia loading on
the beam (i.e. to Zo2 mwðxÞ per unit length) but is in phase with the local velocity. It provides a continuous
input of energy into the beam, counteracting the energy loss due to the damping and maintaining the non-
decaying modal motion.
The question arises as to whether the measured decay rates on the spatially decaying wave of a long, point-
excited beam can yield the same loss factors as those a finite beam resonating in the supposedly spatial-
sinusoidal waves of Ross, Ungar and Kerwin, or in the more complex damped normal modes of Mead &
Markus (M&M). The question is addressed in this section.
The supposed advantage of obtaining loss factors from long-beam spatial decay rate measurements is the
ability to find them at any frequency over a wide range. If there are no reflected waves in the forced motion,
waves of any wavelength can be excited This contrasts with loss factor measurements of the modes of finite
beams which can only be readily obtained at, or close to the widely spaced discrete resonant frequencies. But
this is no problem when one conducts ‘computer experiments’ based on Eq. (13). Continuously changing L in
Eq. (13) leads to a continuous change in the RUK frequency Re(O), the associated modal loss factor Z varying
continuously with it. For each value of Re(O), spatial decay rates of the long-beam, point-excited wave can be
computed from Eqs. (11) or (12) and ‘experimental loss factors’ can be computed.
Some results obtained from this procedure are presented in Figs. 3a,b. These compare the loss factors of the
simply supported beam vibrating in its fundamental damped normal mode with loss factors deduced from
wave decay rates at the same frequencies. Two specific sets of Y’s and b’s are considered as indicated, with g
and b being assumed to remain constant as O varies.3 There is general similarity of form between the pairs of
curves, each one of a pair reaching its maximum value at almost the same frequency. Over the almost straight
‘skirts’ of the curves, the modal values are approximately twice those obtained from decay rates. This is true
for a wide range of g’s, Y’s and b’s, including the low Y values which correspond to thin damping tapes on
thick plates.
Now the frequency ranges of greatest practical interest are those in which loss factors pass through
maximum values, as these are the ranges in which the high damping capacity of sandwich beam or plate
configurations are best exploited. In these ranges the differences between the loss factors of the finite beam and
those deduced from decay rates are most apparent. It has been found from further calculations that if
Y 42:48, the peak values from decay rates are greater than modal values, but are less if Y o2:48. As Y
3
The unreality of this for real damping materials is well known, but the assumption is justified in the present case since the only purpose
is to compare the loss factors found by different methods in systems with identical parameters.
 ARTICLE IN PRESS
752 D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762

Fig. 3. (a,b) The loss factors of a finite simply supported beam vibrating in its fundamental damped normal mode compared with loss
factors deduced from wave decay rates in a long beam.

increases beyond 2.48, these differences increase, and when Y ¼ 100 (a very large value) the decay rate loss
factor is about 10 times the modal value.
These results demonstrate that substantial disagreement exists between spatial decay rate loss factors and
those of the damped normal modes of simply supported beams. The underlying reason is the critical
dependence of the loss factor on both the nature and value of the wavenumber of the wave motion. In the
decay rate ‘experiment’ the motion has a complex wavenumber whereas that of the simply supported beam is
purely imaginary (i.e. a spatially sinusoidal motion). If the beam has other boundary conditions, its damped
normal modes consist of waves with wavenumbers different from either of these. It remains to be seen whether
any of these loss factors agree with loss factors which can be deduced from forced vibration experiments on
finite beams. This will be investigated in the next section through further computer experiments.

4. The loss factors of single-span free–free and encastré sandwich beams

4.1. The general approach

The difficulty of manufacturing real simple supports for a sandwich beam specimen makes experiments on
such beams impracticable but perfectly free or fully fixed (encastré) boundary conditions are relatively easy to
simulate. Accordingly, this section computes the harmonic forced responses of free–free and encastré beams
and analyses them as though they were experimentally measured data. In this way, ‘pseudo-experimental’ loss
factors are determined by some well-known methods and are compared with values obtained from the M&M
damped normal mode theory and the long-beam decay rate method. The responses are computed for beams
with assigned ranges of g, Y and b and over frequency ranges around the fundamental resonances. The next
sections outline the damped normal mode theory for sandwich beams and the theory for their forced vibration
response. The methods used to deduce loss factors from computed responses are also described.
 ARTICLE IN PRESS
D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762 753

4.2. Outline of the damped normal mode and forced response theories

Damped normal modes consist of complex wave pairs, each with a wavenumber given by Eq. (10). Their
transverse displacements in an infinite beam are best expressed in the exponential form wðxÞ ¼ A expðkxÞ but
the hyperbolic form AS sinhðkxÞ þ AC coshðkxÞ is more convenient for finite beams vibrating in damped
normal modes. When the beam is supported symmetrically and the origin for x is Ptaken midway between the
3
supports, its symmetric modes are given by the sum P of the even functions n¼1 A Cn coshðkn xÞ and the
symmetric modes by the sum of the odd functions 3n¼1 ASn sinhðkn xÞ. The total motion in each case must
satisfy the boundary conditions of the beams and this, of course, can only be achieved at discrete resonant
frequencies. For an undamped beam, these are the simple real natural frequencies, but when the beam is
damped they are complex. When their squares are expressed in the form O2 ð1 þ iZÞ as already described, Z is
the loss factor of the damped normal mode and O is the resonance frequency.
O and Z can only be computed by an iterative process which finds the O2 ð1 þ iZÞ roots of a complex
characteristic determinant derived from the boundary conditions at the ends of the beam. Its terms are
functions of both O2 and the complex k’s which themselves are functions of O2. If the uniform beam is
symmetrically supported and the appropriate hyperbolic functions of knx are used for w(x), satisfaction of the
boundary condition at one end of the beam automatically satisfies them at the other. Details relating to the
determinants for free–free and encastré sandwich beams are included in the Appendix which also sets up the
equations for the responses of these beams to single-point harmonic forces. A closed form solution has been
used for this as distinct from a series form of damped normal modes.
In the current exercise, the single exciting force is placed at the centre of the beam and the total generated
motion is split into two parts:

(a) the motion generated by the force as though it were acting on an infinite beam extending indefinitely in
both directions on either side of the force and
(b) the ‘reverberant’ motion in the finite beam generated by reflection of these waves from the actual
boundaries.
P
To the right of the force, (a) is expressed in the exponential form wþ ðxÞ ¼ 3n¼1 A1;n ekn x , the motion to the
left having similar form. The A’s for the right and left-hand wave-sets are found by satisfying the continuity
and equilibrium conditions at the loading point, and from these A’s one easily determines the total ‘infinite
beam’ motion at any point x. In particular, they are found at the finite beam boundary locations.
The
P reverberant motion generated by boundary reflection is best described by the hyperbolic functions, i.e.
by 3n¼1 ACn coshðkn xÞ for the symmetric modes. The ACn’s in this, of course, are different from those defining
the damped normal modes. The total infinite-beam+reverberant motion must satisfy the boundary conditions
at the ends of the beam, which lead to a set of three equations for the three unknown ACn’s. Solving for these,
one then readily computes the total ‘infinite beam’+reverberant wave motion for any point in the beam (see
Ref. [11] for the first presentation of this method for sandwich beams and plates).

4.3. Methods of deriving loss factors from computed or measured responses

These will be described by reference to Fig. 4, which shows the variation with frequency of the real and
imaginary parts of computed responses at the centre of a free–free sandwich beam excited by a central
harmonic force of constant amplitude. The frequency range covered is only that around the fundamental
resonance. The particular core properties used for this diagram were g ¼ 1, Y ¼ 10, b ¼ 1:0 (Y ¼ 10
represents a core which is slightly thinner than one of the (equal) face-plates; b ¼ 1 and g ¼ 1 are typical
values for damped sandwich cores). The modulus of the imaginary part is plotted to allow its whole positive
and negative range to be included on the logarithmic plot.
The real parts of the response are shown in two forms, ‘uncorrected’ and ‘corrected’, the uncorrected part
increasing indefinitely as O ! 0. This is associated with the rigid body translation of the beam which increases
indefinitely because the rigid body acceleration amplitude remains constant as the forcing frequency changes.
In the region of the fundamental resonance frequency the rigid body translation still contributes significantly
 ARTICLE IN PRESS
754 D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762

Fig. 4. The variation with frequency of the real and imaginary parts of the computed responses at the centre of a free–free sandwich beam
excited by a central harmonic force of constant amplitude. g ¼ 1, Y ¼ 10, b ¼ 1:0.

to the total response and distorts the modal response diagrams. These can only be found accurately by a two-
stage correction process, the first of which is to remove the (real) rigid body component from the computed
total. Now the rigid body displacement varies inversely with O2 so its actual values near the resonance can
easily be deduced from its values as O ! 0. The required correction is easily carried out by computer
subtraction but probably could not be done with sufficient accuracy on actual experimental data. The
computed and corrected real part is also shown on Fig. 4 and now bears much closer resemblance to a
conventional highly damped resonance response curve.
Modal properties such as the loss factor are frequently deduced from experimental response data plotted in
the form of Nyquist diagrams on which the real and imaginary parts of the measured or computed modal
response at different frequencies are plotted one against the another. It has long been known [12] that the
Nyquist diagram for a single-degree-of-freedom system with hysteretic damping is a perfect circle centred on
the imaginary axis and ending (as O ! 1) at the origin. Figs. 5a,b show the Nyquist diagrams for the
uncorrected and first-stage corrected computed responses. Only part of the uncorrected diagram is at all
circular. The first-stage correction makes it ‘much more’ circular but it is still not centred on the imaginary
axis. This is due to the presence in the total response of the small and predominantly real contributions from
all the higher modes of vibration which (at the fundamental frequency) are vibrating below their own
resonances. Over the frequency range of primary interest around the circle, it can be assumed that the total
higher-mode contribution remains constant, the value of which must be subtracted from the first-stage
corrected values. It is found by a process of locating the centre of a true circle which best-fits a selected number
of data points (a minimum of three) around the first-stage corrected diagram. From the coordinates of this
centre one finds the constant to be subtracted from each data point of the first-stage corrected circle to move it
to its ‘correct sdof position’, centred on the imaginary axis. Fig. 5c shows the result of this procedure on the
first-stage corrected circle of Fig. 5b. The best-fit circle is not on this but on the computer screen was
indistinguishable from the ‘circle’ shown.
It remains now to deduce the modal loss factor from the stage 2, fully corrected circle. Three methods can be
employed:

(i) From a simple estimate of the Q-factor (the ‘magnification factor’).

(ii) From the half-power frequency bandwidth of the amplitude-response curve.
(iii) From the quotient of the real and imaginary parts of the response.
(iv) From the imaginary part at true resonance, used in an energy input method.

Method (i) utilises very little data from the corrected circle. It assumes that only one mode participates in
the total motion after the rigid body component has been removed and it only requires the amplitudes of
response at a very low frequency and at the resonance frequency. The quotient of these yields the loss factor if
 ARTICLE IN PRESS
D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762 755

Fig. 5. Nyquist diagrams for the uncorrected response and the first- and second-stage corrected responses of Fig. 4.

the system truly has just one degree of freedom. The method can be used for a free–free beam in the current
computer experiment since precise removal of the rigid body response component can be achieved, but is
unlikely to be practicable in a laboratory experiment.
Method (ii) is too well known to require explanation or description. Method (iii) is based upon equations
for the forced harmonic vibration of a simple hysteretically damped system with one degree of freedom, as
follows.
The equation for the harmonic displacement of the system is ½Mo2 þ ðK þ iHÞw ¼ F or f½1  ðo=on Þ2 
þiZgw ¼ F =K. o2n ¼ K=M and Z ¼ H=K is the system loss factor. The real and imaginary parts of w are,
respectively
 
1  ðo=on Þ2 Z
ReðwÞ ¼  2 F =K and ImðwÞ ¼  2 F =K, (14a,b)
1  ðo=on Þ þ Z 2
1  ðo=on Þ2 þ Z2
so
ImðwÞ  
Z¼ 1  ðo=on Þ2 . (15)
ReðwÞ
It appears that Eq. (15) could be used to find the loss factor of the ideal system from values of Re(w) and
Im(w) computed or measured at any frequency. In practice it runs into trouble when o is very close to on as
the value when o ¼ on is indeterminate. At frequencies very close to on the smallest errors in Re(w) or Im(w)
lead to very inaccurate values of Z, but these difficulties are circumvented by the following modified approach.
Rewrite Eq. (15) in the form Z½ReðwÞ=ImðwÞ ¼ ½1  ðo=on Þ2 . Differentiate both sides of this with respect
to o and rearrange to yield
2 o
Z¼ 2
   . (16)
on ðd=doÞ ReðwÞ=ImðwÞ
 ARTICLE IN PRESS
756 D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762

All the terms in this behave sensibly in the important frequency range close to on. Approximate values
of the derivative can be obtained at different frequencies from the discrete measured or computed values of
Re(w) and Im(w). Denote by Xj, the quotient Re(w)/Im(w) at frequency oj, so its derivative in finite difference
form is
dX ðX j  X j1 Þ
¼
do ðoj  oj1 Þ
and the corresponding value of Zj is
oj ðoj  oj1 Þ
Zj  2 . (17)
o2n ðX j  X j1 Þ
The degree of approximation in this depends, of course, on the size of the frequency interval ðoj  oj1 Þ and
on the accuracy of the available Xj’s. Like Eq. (15), Eq. (17) is true for any value of oj provided the system
really has just one degree of freedom. For a real multi-degree of freedom system, satisfactory values of g can
usually be derived only from a restricted frequency range close to on.
The value thus obtained for Z depends on o2n ¼ K=M. on can easily be identified from the fully corrected
Nyquist diagram as the frequency at which the imaginary part of the response vector vanishes. This shows yet
again the need for the two-stage correction of the measured or computed data.
Method (iv), based on the energy Eq. (11) for the loss factor, requires knowledge of both the energy
dissipated ( ¼ the energy input into the system) and the maximum potential energy when the system vibrates
in the required mode. In the laboratory, measurement of the energy input is a common experimental
procedure, being obtained from the product of the amplitude of the exciting force and the imaginary part of
the displacement response at the excitation point. The maximum potential energy is found most readily at the
frequency at which the real part of the driving point response is zero, for then the maximum kinetic and
potential energies are equal. The kinetic energy can be evaluated by length-wise integration of the product of
the known mass distribution and the square of the measured vibration amplitude. However, this laboratory
procedure was not used in the current computational experiment. The exact potential energy was evaluated
from the exact computed modes of forced vibration but for conciseness the detailed procedure is not presented
here.
As method (iv) involves integrated quantities over the whole length of the beam, the loss factors it yields are
less dependent on the location of the response measuring point than those of methods (i), (ii) and (iii).

4.4. Systems subjected to prescribed harmonic motions; motion-excited beams

The previous section relates to systems excited by a variable-frequency force of constant amplitude. Now a
free–free beam, centrally mounted on a relatively massive shaker, can be ‘motion-excited’ at a constant
amplitude of displacement and the loss factor can be deduced from the measured (or computed) frequency-
dependent transmissibility, T(o). This is the ratio of the complex displacement at (say) the tip of the beam to
the enforced central displacement, and at its fundamental resonance becomes the so-called ‘Q-factor’. For the
ideal single degree of freedom system, this is equal to 1=Z.
The resonances occur at the anti-resonance frequencies of the force-excited beam, the lowest of which is
below the fundamental resonance of the force-excited beam. If both methods of excitation are used on the
same beam, a loss factor can be obtained at an additional discrete frequency, but it must be recognised that it
belongs to a different class of flexural mode. The force-excited modes are genuine ‘free–free’ flexural modes
but the motion-excited modes are those of a cantilever of half the overall beam length, its modal displacements
being superimposed upon a rigid body motion.
The transmissibility of the motion-excited beam is easily found from the same equations and the same
computer programme as used for constant force excitation, after dividing the computed force-excited tip
displacement by the computed source displacement. The loss factor can then be deduced from this by the
Q-factor method (not recommended!) or preferably by analysing the Nyquist diagram of T(o) by methods
similar to those already described.
 ARTICLE IN PRESS
D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762 757

Table 1
Comparison of loss factors found by eleven different methods for a sandwich beam with g ¼ 1, Y ¼ 10, b ¼ 1:0

Method Free-free beam Encastré beam

Z On Z On

Forced response, from the Q-factor 0.373 31.2 0.19 24.96

Forced response, from 1/2 power bandwidth 0.41 31.2 0.188 24.96
Forced response, from component quotient 0.397 (20.0)a 0.186 (15.38)a
Forced response, from component derivatives 0.406 31.2 0.186 24.96
Forced response, from energy input 0.386 31.2 0.148 24.96
Damped normal mode theory (M&M) 0.396 31.34 0.186 24.97
From spatial decay rate method, Eq. (12) 0.192 31.34 0.215 24.97
From spatial decay rate method, Eq. (7) 0.294 31.34 0.357 24.97
Transmissibility, from the Q-factor 0.238 20.1 — —
Transmissibility, from 1/2 power bandwidth 0.358 20.1 — —
Transmissibility, from quotient derivative 0.351 20.1 — —
Transmissibility, from energy input 0.365 21.1 — —
a
Arbitrary frequencies; the loss factors deduced at On were indeterminate.

4.5. Loss factors derived from computed responses

The four methods of Section 4.3 have been used to estimate the loss factors of the fundamental modes of
free–free and encastré sandwich beams with g ¼ 1, Y ¼ 10, b ¼ 1:0 (the same parameter values as considered
for the free-free beam of Fig. 4). Loss factors of these modes have also been computed by the M&M theory.
Values, which would be predicted by the two spatial decay rate methods, have been computed for the same
frequencies as the resonance frequencies of the M&M damped normal modes. All these results are compared
in Table 1. Good agreement is seen between most of the loss factors deduced from finite-beam forced vibration
data and those from the damped normal mode theory.
Loss factors were found from transmissibility data by the same four methods as above but the theoretical
loss factor for the corresponding damped normal mode was not determined. It required a larger
computational programme than that for the damped normal modes of free–free or encastré beams and had
not been developed4. The Q-factor method has previously been used in practice to find loss factors of motion-
excited systems, use being made of the relationship Z ¼ 1=Tðon Þ in which on is the frequency at which the real
part of T(o) vanishes. However, when this method is applied to motion-excited continuous systems, it suffers
from the essential dependence of its Z value on the point at which T(o) is measured. The value quoted in Table
1 was derived from the transmissibility of the beam at its tip and is significantly different from the other three
values which were also derived from the tip transmissibility and which all agree well amongst themselves.
When derived from the transmissibility at a point about 1/6th of the beam length from the tip it is found to be
almost equal to the value derived from the energy input which is quite independent of the measuring point.
For quite different reasons, the loss factors deduced by all the other methods may be found to depend on the
response location to a much smaller extent.
The agreement between the values deduced from the forced vibration responses and those of the M&M
damped normal modes is significant. The damped normal modes are complex and are excited by a
hypothetical distributed force distribution. They are not identical to the modes excited by single-point
harmonic forces and although they may be similar, there is no a priori reason why the loss factors of the
differing modes should be as close as they have been found to be. On the other hand, the loss factors deduced
from free-wave spatial decay calculations or measurements are different from all the others by very significant

4
As already stated, the mode of vibration at the transmissibility resonance is actually that of beam of half the total length, free at one
end and ‘clamped-sliding’ at the other. Its mode is neither symmetric nor anti-symmetric and the six boundary conditions to be satisfied
lead to a 6  6 complex determinantal frequency equation, instead of the 3  3 for symmetric modes.
 ARTICLE IN PRESS
758 D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762

Table 2
Comparison of loss factors found by three different methods for free-free damped sandwich beams with different combinations of g, Y and b

g Y b Method Z On

100 10 0.5 Damped normal mode theory (M&M) 0.0394 71.1

Forced response, from quotient derivative 0.0394 71.08
From spatial decay rate, Eq. (7) 0.0761 72.12
10 31.62 1 Damped normal mode theory (M&M) 0.347 95.64
Forced response, from quotient derivative 0.35 95.4
From spatial decay rate, Eq. (7) 0.709 95.64
1.78 0.5 1.5 Damped normal mode theory (M&M) 0.0819 23.41
Forced response, from quotient derivative 0.0823 23.4
From spatial decay rate, Eq. (7) 0.0485 23.41

amounts, sometimes being greater and sometimes less. The spatial decay rate method is therefore seen to be
quite unsuitable for the measurement of modal loss factors of damped sandwich beams.
This conclusion is drawn from results corresponding to just one set of g, Y and b values. Table 2 suggests its
generality by showing the loss factors for free–free sandwich beams with three different combinations of g, Y
and b. In every case, the damped normal mode values agree closely with those deduced from finite-beam
forced vibration data, but they differ widely from those deduced from spatial decay rates. Spatial decay rates
yield a single loss factor for any given frequency, whereas finite sandwich beams which resonate at that
frequency and have the same layer thicknesses and properties have loss factors which depend on the boundary
conditions [4]. This further confirms the above conclusion.

5. Conclusions

Loss factors of damped sandwich beams cannot be reliably deduced from measured spatial decay rates of
flexural waves in long beams and have been shown to differ substantially from the loss factors of resonating
finite sandwich beams. They may be greater or less than the finite beam loss factors even when measured at the
same frequencies.
Computer-experiments on finite beams forced harmonically close to resonance yield loss factors very close
to those predicted by the Mead–Markus theory of complex damped normal modes. This has been confirmed
for both free–free and encastré sandwich beams. The well-known methods of finding loss factors by the 12-
power bandwidth method and by measurement of the energy input at resonance have been shown to be
accurate for even highly damped sandwich beams, and so also have methods which analyse the complex
components of the forced response close to (and even remote from) resonance frequencies. In principle, these
methods can be used to find loss factors from measured responses close to both resonances and anti-resonance
frequencies, although responses actually measured near to anti-resonance frequencies may be insufficiently
accurate for this purpose.
Sandwich beam loss factors depend not only on the frequency, but also on the mode of vibration. Beams (or
plates) which are identical in all respects apart from their lengths and boundary conditions but which resonate
at the same frequencies, have different loss factors. This difference is most pronounced in the fundamental
modes.
Computer experiments on damped sandwich beams are capable of yielding loss factors of multi-figure
accuracy. This, of course, is impossible from actual laboratory data. Real sandwich beams in the laboratory
behave in a notoriously unpredictable manner. Unless extreme precautions are taken they can yield different
loss factors at different times of day, apparently independently of ambient conditions! In spite of this, the
experimenter should use the most reliable methods available to measure the loss factors. The results of this
paper show those in which confidence may be placed.
 ARTICLE IN PRESS
D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762 759

Appendix A. Equations for the frequency response and loss factors of finite damped sandwich beams

Fig. A1 shows the dimensions and loads on the cross-section of a three-layer sandwich beam. The elastic (Young’s)
moduli of the face-plates are E1 and E3. The complex shear modulus of the core is Gð1 þ ibÞ. Its longitudinal elastic
modulus is assumed to be negligible so E 2 ¼ 0. The transverse direct strain in the core is assumed to be zero so the
transverse (flexural) displacement w(x, t) is the same across a given section. The differential equation governing the
flexural wave motion in the sandwich when it vibrates freely at frequency o is then
 
EI T wvi  gð1 þ ibÞð1 þ Y Þwiv  o2 m wii  gð1 þ ibÞw ¼ 0, (A.1)
where EI T ¼ 121
ðE 1 h31 þ E 3 h33 Þ is the sum of the flexural rigidities of the two face-plates when each plate bends
independently with the same w(x).
g is a ‘shear parameter’ and Y the ‘geometric parameter’ defined by
  
0 G 1 1 d 2123 1 1 1
g ¼ þ ; Y¼ þ . (A.2a,b)
h2 E 1 h1 E 3 h3 EI T E 1 h1 E 3 h3
In physical terms Y can be defined by the relationship
Flextural stiffness of the whole beam section when G ¼ 1
ð1 þ Y Þ ¼ .
EI T
Define the non-dimensional frequency O by O2 ¼ o2 mL4 =EI T where L can be any arbitrary length but for a
finite beam is most conveniently taken as the beam length. For an infinite beam it will be taken as unity.
Define also the non-dimensional shear parameter

GL2 1 1
g¼ þ
h2 E 1 h1 E 3 h3
and the non-dimensional wavenumbers ln ¼ kn L. With L ¼ 1 the kL’s of Eq. (10) in the main text are
automatically the wavenumbers per unit length.
Important relationships between the forces and moments acting on the section and the beam displacement w
and its derivatives were first presented in Ref. [3] and are as follows.
The total vertical shear force on a beam section:
EI T  v 
S¼ w þ gð1 þ ibÞð1 þ Y Þw000 þ O2 w0 . (A.3)
gð1 þ ibÞ
The resultant longitudinal force on a face-plate section (equal and opposite in each plate)
 
EI T 1  iv 
P¼ w þ gð1 þ ibÞw00 þ O2 w . (A.4)
gð1 þ ibÞ d 123
The total bending moment acting on the beam section:
 
1 EI T  iv 
M ¼ M 1 þ M 3 þ P1 h2 þ ðh1 þ h3 Þ ¼ w þ gð1 þ ibÞð1 þ Y Þw00 þ O2 w . (A.5)
2 gð1 þ ibÞ

Fig. A.1. Dimensions and loads on the cross-section of a sandwich beam.

 ARTICLE IN PRESS
760 D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762

x ¼ L=2 of the free–free beam are S ¼ 0; M ¼ 0 and P ¼ 0 With

The boundary conditions at the ends P
wðL=2Þ expressed in the form wðL=2Þ ¼ 3n¼1 Acn coshðln =2Þ, these three equations become
X
3   X3
SðL=2Þ ¼ ACn sinhðln =2Þ l5n þ gð1 þ ibÞð1 þ Y Þl3n þ O2 ln ¼ ACn F S ðln ; O2 Þ ¼ 0, (A.6a)
n¼1 n¼1

X
3   X3
PðL=2Þ ¼ ACn cosh ðln =2Þ l4n þ gð1 þ ibÞl2n þ O2 ¼ ACn F P ðln ; O2 Þ, (A.6b)
n¼1 n¼1

X
3   X3
MðL=2Þ ¼ ACn coshðl2 =2Þ l4n þ gð1 þ ibÞð1 þ Y Þl2 þ O2 ¼ ACn F M ðln ; O2 Þ ¼ 0. (A.6c)
n¼1 n¼1

The complex resonance frequency O2 ð1 þ iZÞ of a damped normal mode is found by making, in the first
place, a reasonable guess at its real part and putting Z ¼ 0. Three corresponding l’s are then found from
Eq. (10), followed by a process of iteration to find the accurate value of O2 ð1 þ iZÞ, which satisfies the above
three equations and are the O2 roots of the determinantal equation
 
 F S ðl1 ; O2 Þ; F S ðl2 ; O2 Þ; F S ðv3 ; O2 Þ 
 
 2 2 2 
 F P ðl1 ; O Þ; F P ðl2 ; O Þ; F P ðl3 ; O Þ  ¼ j½F frfr j ¼ 0, (A.7)
 
 F M ðl1 ; O2 Þ; F M ðl2 ; O2 Þ; F M ðl3 ; O2 Þ 

[Ffrfr] signifying that this matrix pertains to the free–free beam.

Now consider the encastré beam, fully fixed at each of its ends. At x ¼ L=2 the two boundary conditions
are simple, w ¼ 0; w0 ¼ 0. These lead to two simple equations for the ACn’s:
X
3 X
3
wðL=2Þ ¼ ACn coshðl2 =2Þ ¼ ACn F w ðln ; O2 Þ ¼ 0, (A.8a)
n¼1 n¼1

X
3 X
3
w0 ðL=2Þ ¼ ACn ln sinhðln =2Þ ¼ ACn F w0 ðln ; O2 Þ ¼ 0. (A.8b)
n¼1 n¼1

The remaining boundary condition relates to the longitudinal displacements u(x) in the face-plates which
are zero at x ¼ L=2. Hence, u1 ðL=2Þ ¼ u3 ðL=2Þ ¼ 0 and in terms of the three constituent waves they yield
  3 ( )
gð1 þ ibÞY EI T X ln L
u1 ðL=2Þ ¼  ACn 2 coshðln =2Þ ¼ 0.
E 1 h1 d 123 n¼1 ln  gð1 þ ibÞ
If this is satisfied, u3 ðL=2Þ is also zero.
The constant term preceding the summation can be dropped to give
( )
X3
ln L X3
u1 ðL=2Þ ¼ ACn 2 coshðln =2Þ ¼ ACn F n ðln ; O2 Þ ¼ 0, (A.8c)
n¼1 ln  gð1 þ ibÞ n¼1

O and Z for the encastré beam are now found from the complex frequency roots of the determinantal equation
 
 F w ðl1 ; O2 Þ; F w ðl2 ; O2 Þ; F w ðl3 ; O2 Þ 
 
 0 2 2 2 
 F w ðl1 ; O Þ; F w0 ðl2 ; O Þ; F w0 ðl3 ; O Þ  ¼ j½F enc j ¼ 0, (A.9)
 
 F u ðl1 ; O2 Þ; F u ðl2 ; O2 Þ; F u ðl3 ; O2 Þ 

[Fenc] signifying that the matrix pertains to the encastré beam.

The above equations and matrices are also involved in the equations for the response of the beams to
a transverse point harmonic force. First of all, the wave field in an infinite beam due to a single force at the
x-origin of the beam ðx ¼ 0Þ must be found. In the positive x-domain of the beam it consists of three damped
 ARTICLE IN PRESS
D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762 761

waves with wavenumbers having negative real parts, i.e.

X
3
wþ ðxÞ ¼ A1;n ekn x , (A.10)
n¼1

the negative sign being included in the index simply to emphasise that these wavenumbers must have negative
real parts.
There are three known boundary conditions at x ¼ 0 which allow the A01 ’s to be found in terms of the
applied harmonic force F eiot . They are dw=ðdxx¼0 Þ ¼ 0, Sð0Þ ¼ F =2 and uð0Þ ¼ 0. One obtains from these:
X
3
A1;n ln ¼ 0, (A.11a)
n¼1

X
3   X3
A1;n l5n þ gð1 þ ibÞð1 þ Y Þl3n þ O2 ln ¼ A1;n F 1;s0 ðln ; O2 Þ ¼ F =2 (A.11b)
n¼1 n¼1

and
( )
X
3
ln L X
3
A1;n ¼ A1;n F 1;u0 ðln; O2 Þ ¼ 0, (A.11c)
n¼1 l2n  gð1 þ ibÞ n¼1

which have the matrix form

2 38 9 8 9
l1 ; l2 ; l3 ; A
< 1;1 >
> = <0>
> =
6F ; 2
F 2
F 1S0 ðl3 ; O2 Þ 7 A
4 1S0 1 ðl O Þ; 1S0 ðl2 ; O Þ; 5 1;2 ¼ ½F 10 fAo g ¼ 1 F =2. (A.12)
> > : >
>
F 1u0 ðl1 ; O2 Þ; F 1u0 ðl2 ; O2 Þ; F 1u0 ðl3 ; O2 Þ : A1;3 ; 0
;

Hence
8 9
<0>
> =
fAo g ¼ ½F 10 1 1 F =2. (A.13)
: >
> ;
0
By analogy with Eq. (A.7), the shear force, face-plate axial force and bending moment due to this ‘infinite’
wave field at x ¼ L=2 can be expressed in the form ½F 1L=2 fA1 g. ½F 1L=2  has exactly the same form as ½F frfr in
Eq. (A.7) but its hyperbolic terms are all replaced by the appropriate exponential terms, exp ln =2.
The reverberant wave field fAREV g caused by reflection of fAo g at the boundaries generates the forces and
moment at x ¼ L=2 given by ½F frfr fAREV g. (½F frfr  as defined in Eq. (A.7)). The total force and moment vector
from the infinite and reverberant wave fields is therefore ½F frfr fAREV g þ ½F aL=2 fAo g. It must vanish at the ends
of the free–free beam, hence
fAREV g ¼ ½F frfr 1 ½F 1L=2 fAo g ¼ ½F frfr 1 ½F 1L=2 ½F 10 1 ½0; 1; 0T F =2. (A.14)
An expression of identical form applies to the encastré beam, but with [Ffrfr] replaced by [Fenc].
The total displacement w(x) at any point in the beam is now given by
3 h
X i
wðxÞ ¼ A1;n eln x=L þ AREV;n cosh ðln x=LÞ . (A.15)
n¼1

References

[1] H. Oberst, K. Frankenfeld, Über die Dämpfung der Biegeschwingungen dünner Bleche durch festhaftende Beläge: I Miteilung aus der
Physikalisch-Technischen Bundesanstalt, Acustica 2 Akustische Beihefte 4 (1952) 181–194.
[2] D. Ross, E.E. Ungar, E.M. Kerwin Jr., Damping of plate flexural vibrations by means of visco-elastic laminae, structural damping,
American Society of Mechanical Engineers (1959) 49–87.
 ARTICLE IN PRESS
762 D.J. Mead / Journal of Sound and Vibration 300 (2007) 744–762

[3] E.M. Kerwin Jr., Damping of flexural waves by a constrained viscoelastic layer, Journal of the Acoustical Society of America 31 (1959)
952–962.
[4] R.A. DiTaranto, Theory of the vibratory bending of a damped sandwich layer in non-sinusoidal modes, Transactions of the American
Society of Mechanical Engineers, Journal of Applied Mechanics 87 (Ser. E) (1965) 881–886.
[5] D.J. Mead, S. Markus, The forced vibration of a three-layer, damped sandwich beam with arbitrary boundary conditions, Journal of
Sound and Vibration 10 (1969) 163–175.
[6] D.J. Mead, S. Markus, Loss factors and resonant frequencies of encastré damped sandwich beams, Journal of Sound and Vibration 12
(1970) 99–112.
[7] E.E. Ungar, E.M. Kerwin Jr., Loss factors of viscoelastic systems in terms of energy concepts, Journal of the Acoustical Society of
America 34 (1962) 954–957.
[8] D.J. Mead, The existence of normal modes of linear systems with arbitrary damping, in: Symposium on Structural Dynamics, Paper
C9, Loughborough University, UK, 1970.
[9] Y.P. Lu, B.E. Douglas, On the forced vibrations of three-layer damped sandwich beams, Journal of Sound and Vibration 32 (1974)
513–516.
[10] D.J. Mead, Passive Vibration Control, Wiley, Chichester, UK, 1998.
[11] D.J. Mead, Y. Yaman, The harmonic response of rectangular sandwich plates with multiple stiffening: a flexural wave analysis,
Journal of Sound and Vibration 145 (1991) 409–428.
[12] C.C. Kennedy, C.D.P. Pancu, Use of vectors in vibration measurement and analysis, Journal of Aeronautical Sciences 14 (1947)
603–605.